From cd64f9b0e7ff5a4f4971c4b6c8b801ceb58b8225 Mon Sep 17 00:00:00 2001
From: Tanguy Pruvot <tanguy.pruvot@gmail.com>
Date: Thu, 17 Sep 2015 23:39:55 +0200
Subject: [PATCH] Add a new cuda-schedule parameter

0: cudaDeviceScheduleAuto
1: cudaDeviceScheduleSpin
2: cudaDeviceScheduleYield
4: cudaDeviceScheduleBlockingSync

Also set the best one (4) for luffa algo by default...
---
 README.txt     |  4 ++++
 ccminer.cpp    | 17 +++++++++++++++--
 cuda.cpp       |  2 ++
 miner.h        |  1 +
 qubit/luffa.cu |  2 ++
 5 files changed, 24 insertions(+), 2 deletions(-)

diff --git a/README.txt b/README.txt
index 9b4548f..393d766 100644
--- a/README.txt
+++ b/README.txt
@@ -106,6 +106,7 @@ its command line interface and options.
 
   -i, --intensity=N[,N] GPU threads per call 8-25 (2^N + F, default: 0=auto)
                         Decimals and multiple values are allowed for fine tuning
+      --cuda-schedule   Set device threads scheduling mode (default: auto)
   -f, --diff-factor     Divide difficulty by this factor (default 1.0)
   -m, --diff-multiplier Multiply difficulty by this value (default 1.0)
       --vote=VOTE       block reward vote (for HeavyCoin)
@@ -225,6 +226,9 @@ features.
 
 >>> RELEASE HISTORY <<<
 
+  Under Dev...    v1.6.7
+                  Add --cuda-schedule parameter
+
   Aug. 28th 2015  v1.6.6
                   Allow to load remote config with curl (-c http://...)
                   Add Lyra2REv2 algo (Vertcoin/Zoom)
diff --git a/ccminer.cpp b/ccminer.cpp
index 2c9713c..d436ab9 100644
--- a/ccminer.cpp
+++ b/ccminer.cpp
@@ -206,6 +206,7 @@ uint32_t device_gpu_clocks[MAX_GPUS] = { 0 };
 uint32_t device_mem_clocks[MAX_GPUS] = { 0 };
 uint32_t device_plimit[MAX_GPUS] = { 0 };
 int8_t device_pstate[MAX_GPUS] = { -1 };
+int opt_cudaschedule = -1;
 static bool opt_keep_clocks = false;
 
 // un-linked to cmdline scrypt options (useless)
@@ -320,6 +321,7 @@ Options:\n\
                         (matching 2nd gt640 in the PC)\n\
   -i  --intensity=N[,N] GPU intensity 8.0-25.0 (default: auto) \n\
                         Decimals are allowed for fine tuning \n\
+      --cuda-schedule   Set device threads scheduling mode (default: auto)\n\
   -f, --diff-factor     Divide difficulty by this factor (default 1.0) \n\
   -m, --diff-multiplier Multiply difficulty by this value (default 1.0) \n\
       --vote=VOTE       block reward vote (for HeavyCoin)\n\
@@ -393,6 +395,7 @@ struct option options[] = {
 	{ "cputest", 0, NULL, 1006 },
 	{ "cpu-affinity", 1, NULL, 1020 },
 	{ "cpu-priority", 1, NULL, 1021 },
+	{ "cuda-schedule", 1, NULL, 1025 },
 	{ "debug", 0, NULL, 'D' },
 	{ "help", 0, NULL, 'h' },
 	{ "intensity", 1, NULL, 'i' },
@@ -2860,6 +2863,9 @@ void parse_arg(int key, char *arg)
 			show_usage_and_exit(1);
 		opt_priority = v;
 		break;
+	case 1025: // cuda-schedule
+		opt_cudaschedule = atoi(arg);
+		break;
 	case 1060: // max-temp
 		d = atof(arg);
 		opt_max_temp = d;
@@ -3322,7 +3328,7 @@ int main(int argc, char *argv[])
 	/* nvml is currently not the best choice on Windows (only in x64) */
 	hnvml = nvml_create();
 	if (hnvml) {
-		bool gpu_reinit = false;
+		bool gpu_reinit = (opt_cudaschedule >= 0); //false
 		cuda_devicenames(); // refresh gpu vendor name
 		applog(LOG_INFO, "NVML GPU monitoring enabled.");
 		for (int n=0; n < opt_n_threads; n++) {
@@ -3332,11 +3338,18 @@ int main(int argc, char *argv[])
 				gpu_reinit = true;
 			if (nvml_set_clocks(hnvml, device_map[n]) == 1)
 				gpu_reinit = true;
-			if (gpu_reinit)
+			if (gpu_reinit) {
 				cuda_reset_device(n, NULL);
+			}
 		}
 	}
 #endif
+	// force reinit to set default device flags
+	if (opt_cudaschedule >= 0 && !hnvml) {
+		for (int n=0; n < opt_n_threads; n++) {
+			cuda_reset_device(n, NULL);
+		}
+	}
 #ifdef WIN32
 	if (!hnvml && nvapi_init() == 0)
 		applog(LOG_INFO, "NVAPI GPU monitoring enabled.");
diff --git a/cuda.cpp b/cuda.cpp
index 4dc9885..679e77f 100644
--- a/cuda.cpp
+++ b/cuda.cpp
@@ -212,6 +212,8 @@ void cuda_reset_device(int thr_id, bool *init)
 			usleep(1000);
 	}
 	cudaDeviceReset();
+	if (opt_cudaschedule >= 0)
+		cudaSetDeviceFlags((unsigned)(opt_cudaschedule & cudaDeviceScheduleMask));
 }
 
 void cudaReportHardwareFailure(int thr_id, cudaError_t err, const char* func)
diff --git a/miner.h b/miner.h
index 3491a54..683a024 100644
--- a/miner.h
+++ b/miner.h
@@ -529,6 +529,7 @@ extern char* device_name[MAX_GPUS];
 extern short device_map[MAX_GPUS];
 extern long  device_sm[MAX_GPUS];
 extern uint32_t gpus_intensity[MAX_GPUS];
+extern int opt_cudaschedule;
 
 #define CL_N    "\x1B[0m"
 #define CL_RED  "\x1B[31m"
diff --git a/qubit/luffa.cu b/qubit/luffa.cu
index 35547c5..e0960c9 100644
--- a/qubit/luffa.cu
+++ b/qubit/luffa.cu
@@ -44,6 +44,8 @@ extern "C" int scanhash_luffa(int thr_id, uint32_t *pdata, const uint32_t *ptarg
 	if (!init[thr_id])
 	{
 		cudaSetDevice(device_map[thr_id]);
+		if (opt_cudaschedule == -1) // to reduce cpu usage...
+			cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
 
 		CUDA_SAFE_CALL(cudaMalloc(&d_hash[thr_id], throughput * 64));